Turnip yellow mosaic virus protease binds ubiquitin suboptimally to fine-tune its deubiquitinase activity

Single-stranded, positive-sense RNA viruses assemble their replication complexes in infected cells from a multidomain replication polyprotein. This polyprotein usually contains at least one protease, the primary function of which is to process the polyprotein into mature proteins. Such proteases also may have other functions in the replication cycle. For instance, cysteine proteases (PRO) frequently double up as ubiquitin hydrolases (DUB), thus interfering with cellular processes critical for virus replication. We previously reported the crystal structures of such a PRO/DUB from Turnip yellow mosaic virus (TYMV) and of its complex with one of its PRO substrates. Here we report the crystal structure of TYMV PRO/DUB in complex with ubiquitin. We find that PRO/DUB recognizes ubiquitin in an unorthodox way: It interacts with the body of ubiquitin through a split recognition motif engaging both the major and the secondary recognition patches of ubiquitin (Ile44 patch and Ile36 patch, respectively, including Leu8, which is part of the two patches). However, the contacts are suboptimal on both sides. Introducing a single-point mutation in TYMV PRO/DUB aimed at improving ubiquitin-binding led to a much more active DUB. Comparison with other PRO/DUBs from other viral families, particularly coronaviruses, suggests that low DUB activities of viral PRO/DUBs may generally be fine-tuned features of interaction with host factors.

Single-stranded, positive-sense RNA viruses assemble their replication complexes in infected cells from a multidomain replication polyprotein. This polyprotein usually contains at least one protease, the primary function of which is to process the polyprotein into mature proteins. Such proteases also may have other functions in the replication cycle. For instance, cysteine proteases (PRO) frequently double up as ubiquitin hydrolases (DUB), thus interfering with cellular processes critical for virus replication. We previously reported the crystal structures of such a PRO/DUB from Turnip yellow mosaic virus (TYMV) and of its complex with one of its PRO substrates. Here we report the crystal structure of TYMV PRO/DUB in complex with ubiquitin. We find that PRO/DUB recognizes ubiquitin in an unorthodox way: It interacts with the body of ubiquitin through a split recognition motif engaging both the major and the secondary recognition patches of ubiquitin (Ile 44 patch and Ile 36 patch, respectively, including Leu 8 , which is part of the two patches). However, the contacts are suboptimal on both sides. Introducing a single-point mutation in TYMV PRO/DUB aimed at improving ubiquitin-binding led to a much more active DUB. Comparison with other PRO/DUBs from other viral families, particularly coronaviruses, suggests that low DUB activities of viral PRO/DUBs may generally be fine-tuned features of interaction with host factors.
Host-pathogen relationships are complex. The outcome of pathogen infection depends on a subtle balance between host immune responses triggered by infection and pathogen replication aimed at promoting propagation. In recent years, ubiquitination and deubiquitination events have emerged as central processes in antiviral mechanisms and viral multiplication (1)(2)(3)(4)(5). Ubiquitination is the conjugation of ubiquitin (Ub), a highly conserved 76-residue protein, to a target protein, through the formation of an isopeptide bond between the C-terminal glycine residue of Ub to a Lys of the target protein (6). Targets of ubiquitination are cellular proteins mostly involved in host immune responses and/or viral proteins (4). In certain cases, ubiquitin-like modifiers such as SUMO, NEDD8, or Ub-like ISG15 (interferon-simulated gene 15) may also be covalently attached to various substrates (7). Substrates are often polyubiquitinated, i.e. a chain of multiple Ub moieties, each linked by an isopeptide bond, is formed. Depending on the linkage type between distal and proximal Ub, the fate of tagged proteins varies, from targeting to proteasome or other degradation pathways for degradation (8) to nonproteolytic events such as interaction with various partners (6). Ubiquitination is a reversible process. Deubiquitination is catalyzed by deubiquitinases (DUBs), which can cleave isopeptide bonds to either trim, degrade, or edit polyUb chains from substrate proteins (7).
Because viruses strictly depend on the host to replicate and spread, they have evolved to circumvent or even hijack for their own advantage the ubiquitin-dependent responses triggered by entry of virus into the cell and subsequent replication (4,9,10). Indeed, a number of viruses have evolved DUBs (11,12), either to counteract antiviral mechanisms or to favor their replication. The targets of viral DUBs can be cellular and/or viral proteins (11). As an example, deubiquitination of cellular proteins by viral DUBs can down-regulate the production of diverse antiviral molecules such as interferons or cytokines and allow viruses to evade host immune responses (12,13). Another example is the deubiquitination of viral proteins by viral DUBs that avoids their targeting to the proteasome, a process that can be viewed as a rescue of these viral proteins. For some viruses an excess of certain viral proteins can be detrimental for viral replication (14,15). These viruses use the deubiquitination step to modulate proteasome-dependent degradation to subtly control the level of the relevant proteins (9). For instance, adjusting the amount of RNA-dependent RNA polymerase (RdRp) may regulate the replication of some RNA viruses such as Sindbis virus (16), Turnip yellow mosaic virus (TYMV) (14,17), or Hepatitis A virus (18).
DUBs are cysteine proteases or metalloproteases and are classified into seven families including two new families that have been recently defined (7,(19)(20)(21). These enzymes can specifically cleave one or several Ub linkage types or display a more general deubiquitinating activity. DUBs encoded by some single-stranded, positive-sense RNA viruses ((1)ssRNA viruses) such as arteriviruses, coronaviruses, picornaviruses, and tymoviruses are actually bifunctional enzymes also responsible for the viral polyprotein maturation through a protease (PRO) activity that cleaves defined peptide bonds (22)(23)(24)(25)(26)(27). The molecular determinants that regulate these dual activities remain largely unknown.
The dual PRO/DUB enzyme encoded by TYMV is a valuable example to address these questions because it is known to tightly regulate the level of RdRp during viral replication (24,28). TYMV encodes an essential 206-kDa replicative polyprotein called 206K, which contains sequence domains indicative of methyltransferase (MT), PRO, NTPase/helicase (HEL), and RNA-dependent RNA polymerase (POL or RdRp) activities. The TYMV PRO domain first cleaves 206K to give rise to an intermediate product called 140K (encompassing the MT, PRO, and HEL domains) and the protein 66K (POL), after which it cleaves the 140K intermediate to release proteins called 98K (MT-PRO) and 42K (HEL) (29)(30)(31). The 66K polymerase is subject to phosphorylation and ubiquitination events triggered by the host, which ultimately target the modified protein to the proteasome where it is degraded (14,32). Because of its DUB activity, the PRO domain of TYMV can counteract such degradation and inhibit 66K degradation (24). The whole process ensures a low level of 66K/POL in infected cells (33), the accumulation of which is deleterious for viral RNA replication (14). Although TYMV 66K is likely to be tagged with Lys 48 -linked polyUb chains and TYMV PRO/DUB is able to process in vitro Lys 48 -and Lys 63 -linked polyUb chain (24), little is known about the type and the composition of polyUb chains attached to the 66K polymerase. In addition, how TYMV PRO/DUB recognizes ubiquitinated 66K is unknown.
The structure of TYMV PRO/DUB (34) has shown that the protein is a DUB from the ovarian tumor (OTU) family (7) that evolved to acquire a PRO function (34). Strikingly, Tymoviridae PRO/DUBs are the only OTU DUBs that lack two elements of the canonical cysteine protease active site displayed by all other OTU DUBs. First, it has only a catalytic dyad (composed of Cys 783 and His 869 ) instead of the typical (Cys-His-Asp/Asn) triad of OTU DUBs. The Asp/Asn residue is replaced by a serine (Ser 871 in TYMV PRO/DUB) that is conserved in the other members of the Tymoviridae family (34). Second, there is no pocket that could constitute the oxyanion hole that is formed during the catalytic mechanism (34). In contrast, Tymoviridae PRO/DUBs display a unique loop (Gly 865 -Pro 866 -Pro 867 ) in close vicinity of the active site (34). We previously concluded that this loop is involved in substrate recognition and contributes to align the side chains of catalytic residues (28). The mobility of this loop therefore would contribute to switching from the PRO activity to the DUB activity. In one of the TYMV PRO/DUB crystal structures, the protein has adventitiously self-assembled into the active form (35), leading to a physiologically relevant PRO/DUB·PRO complex 4 that gives clues to the mechanism of the PRO function of the enzyme. Indeed, this structure provides a snapshot of how the enzyme recognizes the C-terminal extremity of another PRO domain during the PRO;HEL cleavage event, which occurs in the course of polyprotein maturation (30,34).
To better understand the DUB function of the TYMV PRO/ DUB domain, we report its crystal structure in complex with ubiquitin. We supplemented the low resolution of the structure (3.7 Å) with molecular dynamics simulations. We used this modeling approach to further probe the differences in molecular recognition between two of its substrates, i.e. PRO of the PRO;HEL cleavage site and ubiquitin. A structure-guided mutagenesis study identified point mutants with an increased DUB activity, showing that the unusual recognition of Ub by TYMV PRO/DUB is suboptimal. Comparison of this PRO/ DUB-Ub structure with that of the PRO/DUB·PRO complex that occurs during polyprotein processing (34) shows that these unrelated substrates are recognized by largely overlapping recognition surfaces.

Results
Overall structure of the covalent TYMV PRO-Ub complex To solve the crystal structure of a TYMV PRO/DUB·ubiquitin complex and because the affinity of a single module of Ub for the enzyme is low (24,34), we used a modified form of Ub (Ub-VME) in which the C-terminal Gly 76 is substituted with a vinyl methylester function that spontaneously and irreversibly forms a covalent linkage with the catalytic cysteine of DUBs in a Michael addition (36,37). TYMV PRO/DUB and Ub-VME were incubated at 25°C, leading to the formation of a covalent complex as evidenced by SDS-PAGE (Fig. S1A), which was then purified by size-exclusion chromatography (Fig. S1B). Crystals of the protein complex grew in a single drop after 120 days. Only a single crystal showed acceptable diffraction that allowed us to collect data. The structure was solved at 3.7 Å resolution by molecular replacement. The crystallographic asymmetric unit contains two PRO/DUB-Ub complexes, one of which is well-ordered and could be modeled with confidence, except in a few places where density was ambiguous. We complemented this crystallographic model with molecular dynamics simulations that helped to resolve ambiguities and allowed an accurate view of the complex (see below for details). The second complex in the asymmetric unit was modeled from the first and the structure refined with tight noncrystallographic restraints with good statistics (Table 1). We will limit our analysis to the single well-ordered complex composed of chains A (TYMV PRO/DUB, ordered residues 732-876 by polyprotein numbering) and B (Ub-VME, residues 1-76 including the terminal glycyl-vinylmethylester covalently linked to the catalytic Cys 783 ).
The interaction surface of the PRO/DUB-Ub complex measured by PISA server (38) buries 860 Å 2 (11%) and 908 Å 2 (19.5%) of solvent-accessible area for the TYMV PRO/DUB and Ub molecules, respectively, which is on the lower side of the reported values for other DUB·Ub complexes (39)(40)(41). As in these other complexes, the Ub-binding interface of TYMV PRO/DUB can be viewed as two distinct areas (Fig. 1). First, the body of Ub is bound by a surface of TYMV PRO/DUB distant from the PRO/DUB active site and contributed on one side by its N-terminal lobe (residues 732-770) and on the other by the C-terminal lobe (residues 836-876; for a more detailed description of the three lobes, see Ref. 34). Second, the C-terminal extremity of Ub inserts into the TYMV PRO/DUB catalytic cleft between the central lobe (residues 773-835) and the C-terminal lobe.
TYMV PRO/DUB uses two polar loops to simultaneously engage the two major hydrophobic patches on the body of Ub Distant from the active site, the interaction of TYMV PRO/ DUB with Ub appears quite unusual: Ub plugs into a large groove at the surface of TYMV PRO/DUB, so that both of its major recognition patches (the so-called Ile 44 and Ile 36 patches) are bound simultaneously (Fig. 1A). On one side of the groove, the Ile 44 patch is contacted by the Tymoviridae-specific N-terminal lobe (residues 732-772), whereas on the other side, the Ile 36 patch is contacted by the C-terminal lobe that is common to all OTU DUBs.

The Ile 44 patch-interacting site
The side chains of TYMV PRO/DUB Glu 759 and Asn 760 , from the N-terminal lobe of the protein, project directly toward the Ile 44 patch of Ub, composed of residues Ile 44 , Leu 8 , His 68 , and Val 70 (6) ( Fig. 2A). In previous work based on a docking model of the complex and a subsequent mutagenesis study, we suspected the involvement of Glu 759 and Asn 760 in Ub recognition. We hypothesized the presence of a hydrogen bond between Asn 760 and His 68 and a salt bridge between Glu 759 and Lys 6 and/or His 68 (34). Indeed the simultaneous replacement of these two residues by two Gly residues (mutation E759G/ N760G) led to a small but significant decrease of DUB activity in vitro (34). No such interactions are seen in the crystal structure ( Fig. 2A). However, Lys 6 of Ub is engaged in a strong crystal contact with Asp 739 and Thr 741 from a neighboring molecule  (data not shown). This precludes any interaction with TYMV PRO/DUB in the asymmetric unit but does not exclude the existence of such an interaction in solution. Hence, to better understand the interaction network between TYMV PRO/ DUB Glu 759 and Asn 760 residues and the Ile 44 patch of Ub, we performed molecular dynamics simulations of the complex. For the starting model, we made two changes that depart from the crystal structure: we first replaced the C-terminal ubiquitin residue (a Gly substituted with a vinyl methylester group; see above) with an unmodified glycine. Thus, the complexes we simulated mimicked product-bound states, as in our previous structure of a PRO/DUB·PRO complex (34). Second, we modeled residues 727-731 that are not visible in the electron density map, and we acetylated Ser 727 to better model the native state of the TYMV PRO/DUB domain (that is, linked to the rest of the polyprotein by its N terminus). When free from crystal contacts, Lys 6 now points toward the side chain of Glu 759 . However, in two independent 50-ns simulations, the two residues never engaged in the formation of a stable salt bridge. This is readily shown by the distribution of distances between atom Nz of Lys 6 and atom Oe of Glu 759 , which shows only a minor peak at 2.8 Å (Fig. 2B). Instead the simulations confirm a strong involvement in the interface of the aliphatic portions of the Glu 759 , Asn 760 , Thr 761 , and Thr 763 side chains. They pack against the hydrophobic Ile 44 patch of Ub (Fig. 2C, top panel).
The only polar interaction between these polar residues and ubiquitin is a hydrogen bond between Thr 763 and Gln 49 (;50% occupancy) (Fig. 2C, bottom panel).
Although our previous docked model of the TYMV PRO/ DUB·Ub complex suggested the potential involvement of another part of the N-terminal lobe (including Leu 732 and Leu 765 ) in Ub binding (34), the crystal structure now shows these residues lying at the edge of the Ile 44 patch in the vicinity of Ub residues Gln 49 , Glu 51 , and Asp 52 (Fig. S2) and with, at best, a small contribution to the interface. Simulations consistently show that Leu 732 and Leu 765 actually tend to come away from ubiquitin (data not shown). This is in agreement with our previous report that Ala mutations of these residues (mutation L732A/L765A) showed no effect on DUB activity in vitro (34), ruling out their involvement in Ub binding.

The Ile 36 -interacting site
The Ile 36 patch of Ub, the core of which is composed of residues Ile 36 and Leu 71 (6), is positioned against segment 840-847 of TYMV PRO/DUB (Fig. 2D), with Ile 847 also interacting with Ub Leu 8 (see below). Arg 844 is clearly the TYMV PRO/DUB residue closest to Ile 36 , but density for the side chain of Arg 844 fades beyond its Cg. Thus, to obtain a better view of the Arg 844 side chain, we analyzed its behavior in molecular dynamics simulations. The simulations show that the charged end of the Arg 844 side chain is highly mobile and samples a large conformational space, where it finds several defined bound states. Indeed, guanidinium function of Arg 844 alternatively makes a transient salt bridge with the Ub Glu 34 side chain ( Fig. 2E) or hydrogen bonds with the Ub Gln 40 side-chain or main-chain carbonyls of Ub Glu 34 and Gly 35 (data not shown). In contrast, the aliphatic portion of Arg 844 down to Cg remains stably packed against the Ub Ile 36 patch (Fig. 2F). Thus, we arrived at a similar picture as for the region of the Ub Ile 44 patch, with polar or charged residues of TYMV PRO/DUB contacting the hydrophobic patches of Ub almost exclusively by their aliphatic portions.
The Leu 8 -interacting site Leu 8 of Ub is located between the two hydrophobic patches, in a flexible loop that connects the two first a-helices (42,43). This loop can undergo conformational changes, from a "loopout" to a "loop-in" position (44), which in turn enables it to be part of either the Ile 44 or the Ile 36 patch (44, 45) (Fig. 2G). The flexibility of the loop that comprises Leu 8 is now recognized to be important for recognition of ubiquitin-binding proteins (UBPs) (43). In the TYMV PRO/DUB-Ub complex, this loop adopts an intermediate position between the loop-out and loop-in positions, and Leu 8 points toward the bottom of the groove (Fig. 2G). In this region, Ile 847 and Phe 849 of TYMV PRO/DUB make strong hydrophobic contacts with Leu 8 , Thr 9 , Val 70 , Leu 71 , and Leu 73 of Ub. This centrality of Ile 847 in an interaction network based essentially on hydrophobic interactions is consistent with our previous work. Indeed, mutating Ile 847 to Ala, which conserves its apolar properties, has a significant but mild effect on DUB activity, both in vitro and in vivo, whereas addition of a negative charge in this region (mutation I847D) drastically decreased DUB activity (28,34).
In summary, the crystal structure of the covalent complex between TYMV PRO/DUB and Ub, supplemented by molecular dynamics simulations, shows that Ub nestles in a cavity of TYMV PRO/DUB. This binding mode mimics a clamp that holds Ub through hydrophobic interactions made, surprisingly, by TYMV PRO/DUB polar and charged residues, with one side of the clamp formed by the a2-b2 loop containing the Pro 758 -Glu 759 -Asn 760 -Thr 761 -Ala 762 -Thr 763 motif and the other side of the clamp constituted essentially by residues belonging to b3 and b4 strands, centering on Arg 844 in the b3-b4 loop (Figs. 1B and 2H). The bottom of the Ub-binding groove is composed of hydrophobic residues that are part of b-strands b3 and b4. On its side, Ub engages three binding sites simultaneously, i.e. in addition to its C terminus (see below): the two hydrophobic patches centered on Ile 44 and Ile 36 connected by the loop encompassing Leu 8 . Despite this three-part contact, the buried surface is on the small side compared with other viral DUB·Ub complexes.

TYMV DUB activity can be improved by point mutations that affect the atypical binding surface used to contact Ub
To probe the puzzling use of polar residues in TYMV PRO/ DUB to bind the Ub hydrophobic patches and determine the relative contributions of polar and hydrophobic contacts, we produced and assayed several structure-guided point mutants for DUB activity. The activity of the mutant proteins produced in Escherichia coli was measured using the general DUB substrate ubiquitin-7-amino-4-methyl coumarin (Ub-AMC) as described (24,34). In this in vitro test, TYMV PRO/DUB is not saturated, even at the highest Ub-AMC concentrations attainable (34). It is therefore not possible to determine precisely the k cat and K m parameters. Instead, the assay far from saturation allows the measurement of K app that approximates k cat /K m . First, we mutated polar residues that interact with the Ile 44 hydrophobic patch of Ub. Replacement of Glu 759 or Asn 760 by alanine resulted in substantially increased DUB activity compared with the WT enzyme: 137 6 7% and 135 6 8% for E759A and N760A, respectively (Fig. 3A). The charged carboxylate of Glu 759 is thus actually detrimental to DUB activity. This is consistent with structural data that highlight the importance of the apolar portions of the interfacial residues and the flickering nature of polar interactions, such as the Glu 759 -Lys 6 salt bridge. Finally, DUB activity can actually be increased by the removal of polar groups and maintaining only apolar side chains. In contrast, mutation of Thr 763 to alanine showed a slightly decreased DUB activity (89 6 8%; see Fig. 1G), again in accordance with structural and sequence data (Figs. 1B and 2C, bottom panel).
Second, we wanted to better understand the role of Arg 844 side chain, which can not only make van der Waals contacts with the Ile 36 hydrophobic patch of Ub but also can form hydrogen bonds or a salt bridge with several Ub residues (Fig.  2E). We thus replaced Arg 844 by Ala. This mutation led to a dramatic 3-fold increase of DUB activity (320 6 14%; see Fig. 3A), an effect also observed with the double mutant N760A/R844A (344 6 5%; see Fig. 3A). This implies that Arg at position 844 of TYMV PRO/DUB is detrimental to DUB activity. Because this residue is located far away from the active site, it is likely that its polar side chain alters the binding to Ub rather than affects the turnover of the enzyme. The observation that TYMV DUB activity can be substantially improved by point mutations prompted us to model the interaction of the R844A mutant with Ub. We performed molecular dynamics simulations of the complex in the same conditions as for the WT and with the same initial models, albeit with the truncation of the Arg 844 side chain to mimic an alanine. In two independent replicates the complex shifted from its initial conformation to one in which Ala 844 packs against the center of the Ile 36 patch, as exemplified by the new van der Waals contacts of Ala 844 Cb with Ile 36 Cg1 (Fig. 3B). In contrast, the catalytic dyad's dynamics were not affected, as shown by the continued rarity of the activating hydrogen bond between His 869 and Cys 783 (Fig. 3C). These results show how the complex can easily adjust to the much smaller alanine side chain to effectively shield the Ile 36 patch from solvent, without disturbing the active site. They confirm the centrality of the apolar contact between residue 844 and Ub and suggest that the effect of R844A is indeed on ubiquitin binding rather than on enzyme turnover.
Altogether, these results reinforce the conclusion that TYMV PRO/DUB indeed binds both Ile 44 and Ile 36 patches of Ub suboptimally, contributing to a poor DUB activity. They establish that point mutations aimed at improving Ub binding do result in a considerably increased DUB activity.
Binding mode of the C-terminal tail of Ub: how TYMV PRO/ DUB recognizes different C-terminal sequences In addition to interactions that involve the body of Ub via its two conserved hydrophobic patches, a large portion of PRO/ DUB·Ub interacting surface engages the five C-terminal residues of Ub inserted into the catalytic cleft of TYMV PRO/DUB (Fig. 1A). As expected, the C-terminal tail of Ub adopts a b conformation that creates a dense hydrogen-bonding network with TYMV PRO/DUB residues that belong to the substrate-binding site (Fig. 1B). These involve the backbone carbonyl oxygens and amide hydrogens of Leu 822 , Thr 824 , and Ser 868 , and side chains of Thr 824 and Ser 868 (Fig. 4A). The strong electron density in the vicinity of Arg 74 of Ub was difficult to interpret at this resolution, because the side chains of Arg 72 and Arg 42 also point in the same direction. Again, molecular dynamics simulations were helpful in settling this ambiguity. Only the Cb of the three arginines were initially modeled in the crystal structure. Alternate solutions for their side chains were then generated, all consistent with electron density, and simulations were performed. Simulations nicely converged to the same arrangement in that region, no matter the starting point. We kept this solution for refinement of the crystal structure (Fig. 4B). The three arginines all point toward the acidic S5 pocket of TYMV PRO/DUB, comprised of Glu 816 and Glu 825 (34). Arg 72 and Arg 74 of Ub both make salt bridges with Glu 816 and/or Glu 825 , a type of interaction often seen in other complexes involving Ub (37,40,41,(46)(47)(48)(49). The side chain of Asp 39 from Ub also points toward Arg 74 (the two make a stable salt bridge in the simulations), making a ring of acidic residues around the three clustered arginines (Fig. 4B). and of the catalytic dyad (Cys 783 and His 869 ) (C) was investigated by performing molecular dynamics simulations of the product state complex, using WT TYMV PRO/DUB or R844A mutant. The R844A mutant was generated by truncating the Arg side chain at Cb to mimic an alanine. The distances were measured along the same 90 ns in two simulations as in Fig. 2 (WT, red histograms) and along 90 ns in two simulations for R844A (black histograms). B, distance between the side chains of TYMV PRO/DUB residue 844 (Cb atom) and Ub Ile 36 (Cg1 atom). C, distance between TYMV PRO/DUB Cys 783 (Sg atom) and His 869 (Nd1 atom). The minor peak at 3.5 Å signals alignment of the catalytic dyad.
We can now assess how the PRO/DUB catalytic cleft adjusts to different substrates. Indeed, TYMV PRO/DUB recognizes a consensus peptide substrate (K/R)LXG(G/A/S) (corresponding to positions P5-P4-P3-P2-P1), where X is any amino acid, corresponding to the HEL;POL and PRO;HEL cleavage sites (KLNGA; and RLLGS;, respectively) and the C-terminal extremity of Ub (RLRGG;). The requirements for this sequence can be explained by a comparison of the crystal structure of TYMV PRO/DUB-Ub complex with that of the PRO/DUB·-PRO complex with the C-terminal extremity of a PRO from PRO;HEL cleavage site inserted in the catalytic cleft of a PRO enzyme (see Fig. 3 in Ref. 34). Overall, the acidic S5 pocket of TYMV PRO/DUB, composed of residues Glu 816 and Glu 825 , highly conserved in the Tymoviridae family (see sequence alignment in Fig. 1B), always accommodates a basic residue at position P5 (Lys or Arg, see above). In the specific case of Ub, the combination of the TYMV PRO/DUB acidic patch and an acidic residue from Ub (Asp 39 ) perfectly accommodate the three Arg of Ub, two belonging to its C terminus (Arg 72 and Arg 74 ) and one oriented toward its C terminus (Arg 42 ). The strict requirement for a Leu at position P4 is imposed by the hydrophobic S4 pocket, created by residues Val 840 , Ser 842 , Ile 847 , and His 862 (Fig. 4C) rather conserved in homologous PROs (Fig. 1B). The absence of a real S3 pocket leads to a relaxed specificity at the P3 position, accommodating structurally unrelated residues such as Arg, Leu, or Asn. The conserved Gly at position P2 fits in a pocket containing Ser 868 and Phe 870 . These two residues, conserved in the Tymoviridae family (Fig. 1B), constitute the so-called glycine specificity motif, a common feature of alphavirus PROs and PRO/DUBs (50,51). Finally, limited specificity for a small side chain at position P1 is due to the flexible enzyme's Thr 864 -Gly 865 -Pro 866 -Pro 867 -Ser 868 loop, which regulates the constriction of the S1 pocket and consequently substrate specificity and enzymatic activity (28). The GPP motif is a strictly conserved (Fig. 1B) addition to the OTU DUB fold found in the Tymoviridae family.
In conclusion, the C-terminal residues of Ub assume an extended conformation and occupy the catalytic cleft of TYMV PRO/DUB. They do so by creating an intricate network of salt bridges, further strengthened by numerous hydrophobic contacts. The consensus sequence of the C-terminal extremity of PRO, HEL, and Ub, composed of invariant residues (positions P4 and P2), conserved residues (positions P5 and P1) and nonconserved residues (position P3), eventually defines which residues are specificity determinants. These allow TYMV PRO/DUB to discriminate among its substrates. The TYMV PRO/DUB is actually a deubiquitinase that acquired a protease function to process its polyprotein (34) (see also "Discussion"). It is likely that the PRO's substrate cleavage sites have evolved to mimic the C-terminal extremity of Ub. Such optimization of substrate sequences allows a single enzyme to perform several enzymatic reactions. Although this may be a simplification, it supports the "genetic economy" concept that

TYMV PRO/DUB contacts the bodies of unrelated substrates through highly overlapping recognition patches
By comparing the PRO/DUB-Ub and PRO/DUB·PRO complexes (Fig. 5, A and C), we show that the TYMV PRO/DUB recognition surfaces for two of its substrates overlap to a large extent (Fig. 5, B and D). Notably, as for Ub binding (Fig. 5B), PRO binding involves residues Glu 759 -Asn 760 on one side and Arg 844 on the other, with Ile 847 in the middle (Fig. 5B). However, the N terminus of TYMV PRO/DUB differently recognizes the substrates. Although the Pro 733 -Ala 734 -Pro 735 motif is prominently involved in PRO recognition (Fig. 5D), only Leu 732 (Fig. 5B) makes a tenuous contact to Ub in the crystal structure (Fig. S2), a contact that is not stable in simulations (see above). The Pro 733 -Ala 734 -Pro 735 motif provides a strong additional apo-lar contact that makes the PRO/DUB·PRO complex less dependent on the hydrophobic bottom of the binding groove. Indeed, our structural data and mutagenesis studies (this work and Refs. 28 and 34) establish Ile 847 as a central residue for Ub recognition but with less of an impact on PRO binding. In addition, the enzyme harbors a single catalytic site, comprised of Cys 783 and His 869 , for both its protease and deubiquitinase activities. PRO and Ub thus share the same TYMV PRO/DUB ligand-binding site and bind in an orientation that exposes their C-terminal extremity toward the catalytic residues. Their interactions with TYMV PRO/DUB are therefore mutually exclusive and compete for binding to the enzyme. This regulates the dual PRO and DUB activities, both in time (proteolytic maturation of the polyprotein at early stages of infection and then regulation of the 66K RdRp amount in later stages) and in space (within the cytoplasm where the polyprotein is translated; then within the viral replication complexes where the viral RNA genome is replicated).

Discussion
Ub is a small molecule that interacts with many very different partners. Despite the wide variety of structural folds and functions encountered in UBPs, Ub interacts with most of them through the same surface(s). In most of the Ub·UBP complexes, Ub engages a canonical protein interaction site known as the "hydrophobic Ile 44 patch" (6,52). A second hydrophobic patch of Ub, centered around Ile 36 , can also be targeted by UBPs (6). Although the Ile 44 patch is a well-known hot spot used by Ub to interact with its partners, fewer studies report an Ile 36 -based interface (44,53). In addition, growing evidence shows the importance of Leu 8 in UBP binding. Leu 8 is located between the two hydrophobic patches and is usually considered to be part of the Ile 44 patch (6,52). However, Leu 8 is located in a flexible loop that can undergo conformational changes (42,43), shifting from a "loop-out" to a "loop-in" conformation (44). In turn, Leu 8 can be displaced from the Ile 44 patch to become a component of the Ile 36 patch (44).

An unusual mode of ubiquitin binding
In TYMV PRO/DUB-Ub complex, Ub engages not only both of its two hydrophobic patches simultaneously but also the loop that comprises Leu 8 (Fig. 2), a mode of binding without precedent thus far (see below). To score the relative importance of the residues interacting with the two hydrophobic patches, we designed and assayed nonconservative TYMV PRO/DUB mutations aimed at disrupting the binding interface. Mutation of residues that interact with Ile 44 patch (E759A, N760A, and T763A) had a mild effect on DUB activity (Fig. 3A), probably because of their contribution in Ub recognition. Altering the central residue (mutation R844A; Fig. 3A) in the interaction of TYMV PRO/DUB with Ub Ile 36 patch dramatically improved DUB activity, a result that likely reflects improved Ub binding, as confirmed by molecular dynamics simulations. This binding interface thus appears far from optimal for ubiquitin binding. Regarding the motif that interacts with Ub Ile 8 , we had previously shown the critical role played by TYMV PRO/DUB Ile 847 in Ub recognition (28,34). The crystal structure of the TYMV PRO/DUB-Ub complex presented here establishes that Ile 847 and Phe 849 engage in strong hydrophobic contacts with Ub Leu 8 and Thr 9 from the flexible loop (Fig. 2G). In addition, this loop adopts a position where Leu 8 no longer belongs to any hydrophobic patch but instead forms a distinct hydrophobic motif (Fig. 2, G and H). Altogether, our results show that the primary determinant of the TYMV PRO/DUB·Ub interaction is centered neither around Ile 44 as usually observed nor around the Ile 36 patch. Instead, Leu 8 , located between the two hydrophobic patches, directly interacts with TYMV PRO/DUB Ile 847 , located between the two polar patches that sense the Ub patches. The Leu 8 :Ile 847 pair therefore makes the major contribution to this interaction. Leu 8 could thus be a major determinant in Ub involved in sensing its partners (43,44,54).
It is interesting to compare how Ub binds to different viral DUBs, including those that have the dual PRO and DUB activities. The other viral OTU DUBs for which the structures of complexes with Ub are available are encoded by Crimean-Congo hemorrhagic fever orthonairovirus (CCHFV) (55), Dugbe virus (DUGV) (56), and Equine arteritis virus (EAV) (25). The EAV PLP2 is an interesting case. It is an OTU PRO/ DUB like TYMV's. Furthermore EAV belongs to the order Nidovirales that also includes coronaviruses, members of which have caused three deadly epidemics in the 21st century including the current COVID-19 pandemic (57). All Nidovirales encode several proteases, at least one of which is a papain-like protease that doubles up as a DUB (58). However, in coronaviruses this PRO/DUB does not belong to the OTU family as in EAV, but to the ubiquitin-specific protease (USP) family (7). This illustrates the capability of RNA viruses to acquire multiple cellular genes for the same function. It also underlines the major argument in favor of the view that TYMV PRO/DUB is a modified cellular DUB that secondarily acquired its processing protease function: it belongs to a family (OTU) of strict DUBs with no PRO activity, the only exceptions being a few viral OTU DUBs with dual PRO/DUB activity. We include in our structural comparison the USP PRO/DUBs encoded by the coronaviruses SARS-CoV (41), MERS-CoV (59), and MHV (PDB 5WFI). 5 The comparison is also extended to cellular DUBs of the OTU family. We use cellular OTU DUBs from yeast (60) and human (61). The mode of interaction of Ub with TYMV PRO/DUB is thus seen to be very divergent. In all cases, other viral DUBs interact with the body of Ub only through its Ile 44 patch (Fig. 6). The Leu 8 loop of Ub is most often found in the "loop-out" conformation ( Fig. 6, insets), Leu 8 being consequently part of Ile 44 patch, including for cellular OTU DUBs. In the viral complexes with MHV PLP2, EAV PLP2 or CCHFV vOTU, the Ub Leu 8 loop occupies the intermediate position observed in TYMV PRO/DUB-Ub complex (Fig. 6, insets). The loop adopts this intermediate position in all other crystal structures of complexes involving CCHFV vOTU and Ub (62, 63) (data not shown). These comparisons show that Ub Leu 8 usually belongs to the Ile 44 patch but also can be located between the two Ub hydrophobic patches to contact its partner. This intermediate position is found regardless of the enzyme considered, i.e. either a dual PRO/DUB or a DUB, either of viral or of cellular origin. Therefore, the function of Leu 8 is not a hallmark of a DUB family but a specific feature of some enzymes, such as TYMV PRO/DUB.
Superimpositions of TYMV PRO/DUB with cellular OTU DUBs show that yeast OTU1 and human OTUD2 also interact simultaneously with the two hydrophobic patches of Ub (Fig.  S3A) but engage mainly hydrophobic residues, together with one charged residue that structurally resembles Arg 844 of TYMV PRO/DUB, i.e. Glu 203 in yOTU1 or Arg 245 in hOTUD2 (Fig. S3). From an evolutionary point of view, TYMV PRO/DUB appears to be a cellular OTU DUB that has acquired a PRO function by retaining the clamp that holds Ub but losing important hydrophobic residues that interact with the two hydrophobic patches of Ub. This produces an enzyme with low DUB activity.
The low DUB activity of TYMV PRO/DUB may be an evolutionary compromise that ensures proper viral replication TYMV PRO/DUB exhibits a significant but low deubiquitinase activity, its catalytic efficiency k app (which approximates k cat /K m ) being ;2.5 3 10 3 M 21 s 21 (24,28,34), which is 10-1,000-fold lower than that of other DUBs such as CCHFV vOTU, EAV PLP2, or MERS-CoV or SARS-CoV PL pro (25, 55,56,64). Several nonmutually exclusive hypotheses can be proposed that explain this low activity.
First, the crystal structure of TYMV PRO/DUB showed that the protein structurally belongs to the OTU superfamily of DUBs but displays a peculiar active site. Indeed, TYMV PRO/ DUB has no Asp or Asn that are usually part of the catalytic triad of OTU DUBs in combination with a Cys and a His, nor any oxyanion hole to stabilize the thioester intermediate of the catalytic mechanism (34). This results in an altered active site and explains the low DUB activity. In some cases, one or several catalytic functional groups are provided by the substrate, restoring a functional active site, a phenomenon called sub-strate-assisted catalysis (65). The crystal structure of the TYMV PRO/DUB-Ub complex presented in this work shows that Ub does not supply any residue that would restore a complete DUB active site. However, because Ub is located in the P side of TYMV PRO/DUB, it cannot be ruled out that the third residue may be provided by the substrate positioned in the P9 side of the enzyme, i.e. the polyubiquitinated polymerase.
Second, the TYMV DUB activity is measured in vitro with a recombinant PRO/DUB domain and a single Ub molecule, whereas in vivo the enzyme is present in large macromolecular assemblies. Many deubiquitinases possess additional domains, built around a structurally conserved DUB scaffold, that are involved in substrate specificity and regulation of DUB activity. Additional protein domains of the TYMV replication protein may interact with the PRO/DUB domain and/or with its  (59)), MHV (PDB code 5WFI, unpublished structure), and EAV (PDB code 4IUM (25)) (black) and DUBs encoded by CCHFV (PDB code 3PHW (55)) and DUGV (PDB code 4HXD (56)) (blue). We also compared DUBs from yeast (PDB code 3BY4 (60)) and human (PDB code 4BOS (61)) (magenta).
substrate and thus contribute to the regulation of TYMV DUB activity. DUB activity in vivo is in fact carried by the 98K protein (24), a large multidomain protein that comprises both the MT and PRO/DUB domains, separated by a region harboring the chloroplast-targeting domain (66) and a proline-rich region. This domain organization is similar to that found in the C-terminal part of nsP2 protein of numerous alphaviruses. This consists of an N-terminal protease subdomain (nsP2pro) and a C-terminal subdomain with a methyltransferase fold (MT-like), connected by a long loop. Crystal structures of the C-terminal part of nsP2 protein of Chikungunya virus (CHIKV) and Venezuelan equine encephalitis virus (VEEV) show several intramolecular interactions between the two subdomains and the involvement of the MT-like subdomain in nsP2pro function (51,67,68). Indeed, the nsP2pro active site is located at the interface between the two subdomains, and its accessibility is regulated by the interdomain loop. Moreover, the MT-like subdomain actively participates in nsP2pro's substrate recognition and binding. Although the nsP2pro subdomains of CHIKV and VEEV do not display DUB activity, these findings illustrate the opportunities for protease regulation inherent to the inclusion of the TYMV PRO/DUB domain in a larger protein. Crystal structures comprising full-length polyubiquitinated 66K and/ or 98K protein should help to understand the role of the substrate and/or the other domains of 98K protein in TYMV DUB activity.
Third, Arg 844 may contribute to the protease activities of TYMV PRO/DUB because it forms a minor contact to the substrate in the PRO/DUB·PRO complex (Fig. 5, C and D). Analyses of the PRO;HEL and HEL;POL cleavages in vivo indicate that processing of the polyprotein is not affected by the R844A mutation (Fig. S5). However, we cannot rule out that proteolytic activity also occurs on presently unknown cellular substrates, as with other viral proteases. In such a case, the presence of the nonoptimal Arg 844 to contact Ub could be a tradeoff to bind efficiently other substrates with unrelated surfaces.
Fourth, and this is our preferred hypothesis in view of the lack of discernible effect on processing protease activity of R844A, low DUB activity may be a fine-tuned feature of TYMV PRO/DUB. Indeed, viral proteases are usually highly specific enzymes whose activity depends not only on the particular sequence of a cleavage site but also on the remainder of the substrate. The cleavage site, often located in a solvent-exposed flexible loop, is commonly recognized by proteases in an extended conformation that favors its perfect positioning into the catalytic cleft (69). Substrate specificity is ensured by specific interactions between the body of the substrate and the enzyme. Although Ub is usually recognized through its Ile 44 hydrophobic patch only (6,52), we show in this work that TYMV PRO/DUB, unlike other viral PRO/DUBs, has maintained the cellular OTU DUB mode of recognition, involving the two hydrophobic patches on Ub simultaneously. Nevertheless, the interacting surface is both small and suboptimal in its composition, as shown by mutants that improve the DUB activity (Fig. 3A). Because recognition of the substrate body is usually the driving force that allows enzyme/substrate recognition, this observation is puzzling. Residues involved in Ub recogni-tion are not conserved in the Tymoviridae family ( Fig. 1B and  Fig. S4). We hypothesize that maintaining a low DUB activity may be an evolutionary compromise to ensure proper viral replication. Indeed, although it is initially produced in amounts equimolar to 98K and 42K (the two other products of the 206K polyprotein maturation process), 66K displays a transient accumulation in the viral replication cycle (33). The 66K polypeptide is degraded at a late stage of viral infection by the ubiquitin proteasome system through polyubiquitination (14). Nevertheless, by harboring a DUB activity, TYMV possesses a rescue system to avoid complete degradation of the 66K protein. This ensures maintenance of the appropriate level of polymerase and safeguards efficient replication of the TYMV genome. This level may be reached with low DUB activity. In the case of TYMV, too high a level of 66K is actually detrimental to replication. We suggest that a finely tuned DUB activity may be a general feature of viruses that use deubiquitination to adjust the amount of protein(s) that is/are critical for their replication. Indeed, Lei and Hilgenfeld (40) found that the MERS-CoV PL pro Ub-binding surface is likewise suboptimal and nicely discussed the functional implications of this finding. Future experiments will be aimed at determining whether this applies also to the case of TYMV.

Materials and methods
Covalent coupling of Ub to TYMV PRO/DUB TYMV recombinant PRO/DUB fused to an N-terminal His 6 tag (34) was expressed and purified as previously described (28), and diluted to a final concentration of 5 mg/ml in a fresh buffer composed of 10 mM Tris-HCl, 350 mM ammonium acetate, 1 mM DTT, pH 8. A C-terminally modified vinyl methyl ester variant of HA-tagged ubiquitin (HA-Ub-VME) was prepared in 50 mM sodium acetate, pH 4.5, essentially as previously described (36). To adjust the pH, HA-Ub-VME was then diluted 10-fold in binding buffer (50 mM Tris-HCl, 150 mM NaCl, pH 8) and incubated for 10 min at 25°C. Conjugation of both proteins was achieved by adding a 2-fold molar excess of His 6 -PRO/DUB to HA-Ub-VME followed by incubation at 25°C for 30 min. Unreacted proteins were removed by sizeexclusion chromatography on a HiLoad 16/600 Superdex 75pg column (GE Healthcare) with 50 mM Tris-HCl, 500 mM NaCl, pH 8, as elution buffer. Elution fractions were verified by 16.5% Tris-Tricine SDS-PAGE, and those containing pure 6His-PRO/DUB-HA-Ub complex were pooled and dialyzed overnight at 4°C against binding buffer. Covalent complex was then concentrated to 24 mg/ml using ultrafiltration on Amicon Ultra with a cutoff of 10 kDa and frozen in liquid nitrogen for storage at 280°C.

Crystallization of His 6 -PRO/DUB-HA-Ub complex
Crystallization conditions of His 6 -PRO/DUB-HA-Ub complex were screened by a robot using commercial kits from QIAgen and the sitting-drop vapor-diffusion method. Some promising conditions were manually reproduced at 19°C in larger drop volumes (1 ml of 15 mg/ml complex solution plus 1 ml of crystallization reagent equilibrated against a 0.5-ml reservoir volume) using the hanging-drop vapor-diffusion setup. Few Suboptimal binding of ubiquitin by TYMV PRO/DUB crystals appeared after several months in 20% PEG-20K, 0.1 M MES-NaOH, pH 6.5. Prior to data collection, these crystals were harvested, transferred to a cryo-protectant solution (21% PEG-20K, 0.1 M MES-NaOH, pH 6.5, 20% glycerol), and flashfrozen in liquid nitrogen.

Data collection and processing and structure determination
Data collection was performed at Beamline PROXIMA-1 at French synchrotron SOLEIL. Only one crystal showed correct diffraction, and a complete data set could be collected at 3.66 Å. The data were processed and scaled with XDS (70). Because structures of individual TYMV PRO/DUB and human ubiquitin were available, the structure of the His 6 -PRO/DUB-HA-Ub complex was solved by molecular replacement. Calculation of the Matthews coefficient (71) suggested two complex molecules in the asymmetric unit, and several molecular replacement protocols were tested with Phaser (72). The good solution consisted of first locating one complex molecule using a C-terminally truncated version of TYMV PRO/DUB (PDB code 5LW5 chain A (28)) and ubiquitin (PDB code 1UBQ (73)) as search models and second using the resulting solution as an input to find the second complex molecule. The electron density was of sufficient quality to manually rebuild the model in COOT (74). Initial stages of refinement were done with REFMAC (75) and then with PHENIX (76). Because of the low resolution, no solvent molecules could be modeled. The final model thus consists of two TYMV PRO/DUB molecules (residues 732-876 in chain A and residues 732-876 in chain C; His tags could not be modeled) and two HA-Ub-VME molecules (residues 1-76 in chain B and residues 1-76 in chain D; HA tags could not be modeled). The data processing and refinement statistics are listed in Table 1.

Molecular dynamics simulations and structure visualization
Molecular dynamics simulations of a TYMV PRO/DUB·Ub product state complex and of a R844A mutant thereof were performed using the AMBER16 program suite (77) with the ff14SB force field. We noted in preliminary simulations comprising residues 732-876 of PRO/DUB that the first residues tended to interact with Ub, but this seemed to be influenced by the 11 charge spuriously added to Leu 732 by taking it as the N terminus. Thus, we simulated a complex made of an N-acetylated PRO/DUB 727-876 (residues 727-731 were modeled stereochemically) and all ubiquitin residues (1-76). The LEaP program was used for preparation of the systems. Hydrogen atoms were added with default parameters. The complexes were neutralized with K 1 cations and immersed in an explicit TIP3P water box with a solvation shell at least 12 Å deep. The systems were then minimized and used to initiate molecular dynamics. All simulations were performed in the isothermal isobaric ensemble (p = 1 atm, T = 300 K), regulated with the Berendsen barostat and thermostat (78), using periodic boundary conditions and Ewald sums for treating long range electrostatic interactions (79). The hydrogen atoms were constrained to the equilibrium bond length using the SHAKE algorithm (80). A 2-fs time step for the integration of Newton's equations was used. The nonbonded cutoff radius of 10 Å was used. All simulations were run with the SANDER module of the AMBER package. Each complex was simulated for 50 ns twice, and the trajectories were sampled every 10 ps. Analysis of the trajectories with cpptraj showed convergence within the first 5 ns as judged by stabilization of root-mean square deviation. The last 45 ns were kept for analyses.
All simulation trajectories and crystal structures were visualized and structural figures were made with PyMOL (81). PyMOL was also used to mutate Arg 844 to Ala prior to system preparation.

Deubiquitination assay in vitro
Point mutations were introduced in the bacterial vector encoding TYMV PRO/DUB (34) by using the QuikChange II site-directed mutagenesis (Agilent) strategy. Recombinant PRO/DUB proteins were produced and purified as described previously (34). Prior to deubiquitination assay, the purified proteins were dialyzed overnight at 4°C in buffer 50 mM HEPES-KOH, 150 mM KCl, 1 mM DTT, 10% glycerol, pH 8.0, adjusted to a concentration of 100 mM and kept at 280°C until use. The fluorogenic substrate Ub-AMC (Boston Biochem) dissolved in DMSO was diluted in assay buffer (50 mM HEPES-KOH, 10 mM KCl, 0.5 mM EDTA, 5 mM DTT, 0.5% Nonidet P-40, pH 7.8). DUB activity was assessed at room temperature in a Hitachi F2000 spectrofluorometer in assay buffer with a final concentration of DMSO adjusted to 2% to match the DMSO concentration in the highest Ub-AMC concentration assays. Reactions were initiated by the addition of enzyme to the cuvette, and the rate of substrate hydrolysis was determined by monitoring AMC-released fluorescence at 440 nm (excitation at 380 nm) for 10 min. Enzyme concentrations were 125 nM for WT PRO/DUB and mutants. To determine the apparent k cat / K m (K app ), the substrate concentration was kept at a concentration below 0.5 mM with the initial velocity linear in substrate concentration, and K app values were then determined according to the equation V/[E] = K app /[S] as described previously (24). Depending on the batch of Ub-AMC, the DUB activity of the WT enzyme displayed variability, with K app varying between 2,388 6 398 and 2,824 6 213 M 21 s 21 . Hence, the activity of the WT protein was measured as a reference for each independent experiment, and the K app values of mutant proteins were normalized to that of the WT protein measured simultaneously. All experiments were performed at least in duplicate, and the data are expressed as the means and standard deviations of these independent experiments.

Data availability
The structure presented in this article has been deposited in the Protein Data Bank with the following code: 6YPT. All remaining data are contained within the article.
We acknowledge the help from the staff of the Institute for Integrative Biology of the Cell computing facility Service Informatique et Calcul Scientifique (SICS) for the molecular dynamics simulations. Funding and additional information-This work has benefited from the Core Institute for Integrative Biology of the Cell crystallization platform, supported by French Infrastructure for Integrated Structural Biology Grant ANR-10-INSB-05-01. This work and M. A. were supported by the Agence Nationale de la Recherche Contracts ANR-11-BSV8-011 "Ubi-or-not-ubi" and ANR-16-CE20-0015 "ViroDUB." Conflict of interest-The authors declare that they have no conflicts of interest with the contents of this article.